Effects of seasonal adjustment

on the money stock

Robert D. Laurent

The emphasis placed on money stock changes
in the conduct and analysis of monetary
policy has recently increased interest in the
seasonal adjustment of the money stock. This
new interest can be seen several ways. Money
market analysts, for example, are using an ap-
parent seasonal pattern in the seasonally ad-
justed money stock as an aid to prediction.
Also, annual revisions in the seasonal adjust-
ment factors the Federal Reserve uses in ad-
justing raw money figures seem to be getting
bigger. The most recent revisions were the
largest ever.
These developments suggest there could
be problems with the current procedure for
seasonally adjusting the money stock. As a
result, the Federal Reserve has appointed a
commission of outside economists and stat-
isticians to review the seasonal adjustment
techniques used by the Board of Governors in
adjusting financial data.
This article examines the effect of
end holiday buying. Housing starts in March
reflect the beginning of favorable weather for
building. Seasonal adjustments are based on
an assumption that such influences are
regular enough to be represented by calen-
dar months.
The seasonal pattern
Even casual examination of changes in
the money stock shows a pronounced
seasonal pattern. This is true regardless of the
measure of the money stock. But it is most
pronounced for the narrowest measure—M1
(currency plus demand deposits)—on which
this article focuses.
Unadjusted money data show
a seasonal pattern
monthly percentage change in Ml *
4—

seasonal adjustment on the money stock and

3
discusses some of the problems in seasonally
adjusting this series. Though the article con-
2
centrates on the money stock, much of the
discussion applies to any series in which the
1
adjustments for seasonal variations are large.
Data are seasonally adjusted to reveal un-
derlying trends. Retail sales always increase in 0
December, and housing starts always increase
in March. Without adjustment for these
patterns, changes can be deceiving, creating
2
an impression that the underlying trend in
retail sales or housing starts has suddenly
accelerated. Removal from the data of "nor- 3
mal" changes expected during these months
reveals the underlying trend. 4

Calendar months serve as proxies for

5 111111111[11
contemporaneous changes in economic F MA MJ A
1 SON D
forces. Retail sales in December reflect year- Data covers 1968-1977 inclusive.

12 Economic Perspectives
 The money stock contracts for the first
two months of the year, and then expands
through the April tax date. It contracts in May,
then expands through June and July. August
shows a mild contraction before an expansion
that culminates in the year-end holiday
season.
The seasonal adjustment of money is
complex. A simplified description of the ad-
justment is useful in explaining the effects—
and problems—of the adjustment process. 1
The new money-stock figure published every
month is adjusted by dividing previously
determined seasonal adjustment factors into
the raw (unadjusted) figures to obtain the
seasonally adjusted money stock. Thus, a
seasonal adjustment factor of 1.02 indicates
the effect of that particular month is expected
to raise the unadjusted data 2 percent above
the annual average.
The predetermined seasonal adjustment
factors used to adjust data when they are first
published are the most current set of
factors—usually those used to adjust the
previous year's money stock. The seasonal ad-
justment factor used to adjust the first-
published figure for November 1978, say, will
be the same factor used in the 1978 revision of
the November 1977 figure. The factor used to
adjust a particular month's figure is revised
annually for several years after the first
published figure appears. In this article, the
first published seasonally adjusted figure for a
particular month is taken to be the first figure
for that month appearing in the Federal
Reserve Bulletin.
In the adjustment of Ml, separate
seasonal adjustments are made to currency
and demand deposits and the two com-
ponents are then combined to obtain the
seasonally adjusted money stock. Unadjusted
monthly figures for each component are
'For a more comprehensive description of the
seasonal adjustment process, see Thomas A. Lawler,
"Seasonal Adjustment of the Money Stock: Problems and
Policy Implications," Economic Review, vol. 63: no. 6, p.
23, (November-December, 1977), Federal Reserve Bank
of Richmond. This article and its companion article dis-
cuss some of the issues raised here and list references to
other works on the same topic.
Federal Reserve Bank of Chicago 13
divided by the average monthly figure for the
year surrounding that month. The resulting
series of monthly figures, called the seasonal
irregular ratios, shows whether the compo-
nent tends to be high or low that month. The
high level of demand deposits in December,
for example, is reflected in seasonal irregular
ratios that from 1968 through 1977 were
always above 1.02. Over that period,
December demand deposits were always at
least 102 percent of the surrounding 12-
month average.
Monthly seasonal irregular ratios form
the base for the seasonal adjustment factors.
Since the seasonal irregular ratios are ob-
tained by dividing the unadjusted monthly
figure by the average for the surrounding 12
months, there is an implicit adjustment for
growth in the money stock. Also, the seasonal
irregular ratio is based on an assumption that
the effects of different months can be ex-
pressed as percentage changes in the unad-
justed figure. If seasonal effects were assumed
constant—regardless of the size of the money
stock—then the 12-month surrounding
average would be subtracted from the unad-
justed figure.
The effect of changing seasonals
Seasonal adjustment factors can be ob-
tained from seasonal irregular ratios in several
ways. One consideration is whether the
seasonal adjustment factors are assumed con-
stant or whether they are assumed to change
over time. If they are assumed constant, equal
weight is given to the same month every year.
If the factors are believed to change over
time, more weight is given to seasonal
irregular ratios in nearby years.
Current seasonal adjustment of the
money stock assumes that seasonal factors are
changing. The rationale for a moving seasonal
is that the economic forces for which the
seasonals serve as proxy may be changing.
The importance of some holidays—or vaca-
tion habits—may change slowly over time,
changing seasonal adjustment factors.
The extent to which the seasonal for any
calendar month is allowed to vary over
different years depends critically on the
relative weight given to the seasonal irregular
ratios for that month in determining the
seasonal adjustment factor. The more weight
given the current year's seasonal irregular
ratio, the more seasonal adjustment factors
are allowed to vary from year to year.
At one extreme, every year's seasonal
irregular ratio is given the same weight and
the seasonal adjustment factors are equal for
every year. The seasonal adjustment factor for
a particular calendar month can, of course,
change over time with the addition of new in-
formation. At the other extreme, the seasonal
irregular ratio is given all the weight, making
the seasonal adjustment factor identical to the
seasonal irregular ratio in that month.
The first extreme, having constant
seasonals, produces a fairly volatile money
stock. The second, with more volatile
seasonals, produces a money stock that
moves smoothly with changes in trend
growth.
In practice, each seasonal adjustment
factor is determined by a seven-year span of
seasonal irregulars centered on that year, the
weights declining as data move away from the
center year. Thus, the November 1978
seasonal adjustment factor will eventually be
determined by November seasonal irregular
ratios from 1975 through 1981. This procedure
allows seasonal adjustment factors to vary
significantly year to year, since the three mid-
dle years of the seven-year span are each
given a fifth of the total weight in determining
the seasonal adjustment.
Seasonal adjustment factors, however,
are not obtained simply by the mechanical
manipulation of the seasonal irregular ratios.
Extreme seasonal irregular ratios are given
less weight in determining seasonal adjust-
ment factors. This provides a more stable
seasonal adjustment factor series and
prevents an inordinate response to extreme
changes in the data. More important,
seasonal adjustment factors are adjusted
judgmentally to take account of other factors.
Often of significant magnitude, these
judgmental adjustments can take account of
sharply changing factors that, while known to
14 Economic Perspectives
affect the money stock, would not be picked
up by a moving seasonal. Examples are
monetary policy changes and changes in tax
dates and holidays.
The smoothing effect of seasonal revisions
The process used to adjust money-stock
data produces large subsequent revisions that
smooth out the first-published figures. When
the money stock figure is first published, the
seasonal adjustment factor used for the
month gives no weight to the unadjusted
money stock for that month. Later revisions,
however, use that month's seasonal irregular
ratio in determining the seasonal adjustment
factor. As a result, revisions move the seasonal
adjustment factor closer to the actual seasonal
irregular ratio. This has the effect of smooth-
ing seasonally adjusted changes in the money
stock.
Although extremes of the seasonal
irregular ratio are given reduced weight in the
determination of seasonal adjustment factors,
the smoothing effect of seasonal revisions on
extreme changes in the money stock is
pronounced. From 1968 to 1977, there were 17
instances where the first published data
showed monthly increases in the money stock
of more than 1 percent and 16 instances
where the data showed declines. In the 1978
number of observations
The first published MI figures ...
16
14
12
0
8
6
4
0 I 11111111 1111111111111
monthly percentage change in M1'
-1.0 .9 .8 .7 .6 .5 .4 .3 .2 .1 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1.0 1.1 1,2 1.3 1.4 1.5 1.6 1.74
. Ddld covers 1968-1977 inclusive.
revision, every one of the 17 large increases
had been reduced and 15 of the 16 declines
showed smaller declines. Only seven in-
stances of money declines remain in the most
recent revision.
Two factors account for the difference
between the first-published data and later
revisions. One is changes in the underlying
raw data. The other is changes in seasonal ad-
justment factors. Separation of the effects of
these two factors make it clear that almost all
the smoothing comes from seasonal revisions.
From 1968 through 1977, the first-
published data show a mean monthly change
in money of 0.461 percent. A measure of the
dispersion in the data is the average absolute
deviation from this mean value—the average
difference (up or down) between each obser-
vation and the average observation. The
average absolute deviation of first-published
data is 0.376 percentage point. When the first-
published data are adjusted to reflect
revisions in the raw data, the mean monthly
change rises to 0.494 percent and the average
absolute deviation falls slightly to 0.364
percentage point. When these data are
further adjusted to reflect the 1978 revisions
in seasonal adjustment, the mean monthly
change becomes 0.490 percent and the
average absolute deviation falls to 0.267
percentage point.
Clearly, revisions in the seasonal adjust-
ment sharply reduce the dispersion of the
monthly percentage changes in the money
supply—and the revisions are significant. The
average absolute change produced by
seasonal revisions was 0.220 percentage
point—nearly half the mean percentage
change before seasonal revision.
A problem for analysis
The pronounced smoothing that comes
from seasonal revisions raises several
questions about the seasonally adjusted
money series. Aside from whether the first-
published data or the later revisions reflect
seasonal adjustments more accurately,
smoothing poses problems of consistency for
analysts of monetary policy.
Most research that relates changes in the
money stock to changes in economic activity
uses the smoothed seasonally adjusted
money series. Though seasonal revisions do
not usually change the general pattern of
acceleration or deceleration in growth of the
money stock, the revised series is con-
siderably smoother than the first-published
data. By relating this smoothed series to
variations in economic activity, analysts

. . . are not greatly affected by revisions in . but are considerably smoothed by revisions
the underlying raw data .. . in the seasonal adjustment factors
number of observations number of observations

16 — 16 —

14 14

12 — 12

10 — 10

8 8

.54131 1 1 1 , 6 1 91,1 ,1 „ 1,
6 6

4 4—

2 — 2 L

01 o_
—10 .9 .1

monthly percentage change in M1. monthly percentage change in M1"
• Data covers 1968-1977 inclusive. •Data covers 1968-1977 inclusive.

Federal Reserve Bank of Chicago 15

associate dampened accelerations or
decelerations in money with fluctuations in
economic activity. As a result, even small
changes in money growth seem to have fairly
large effects on economic activity. But the
money data used in analyzing current policy
are the volatile first-published figures. In-
serted into the historical relationship
between money and income, these un-
smoothed data imply unduly large changes in
economic activity.
As an example of the smoothing process,
the first-published M1 data for February 1971
through January 1972 showed a six-month
period in which the money stock grew at an
annual rate of 12.5 percent followed by six
months in which the annual rate was 0.9 per-
cent. Recent revisions of the data show,
however, that the money stock grew at a rate
of 9.2 percent in the first six months and 4.4
percent in the second. Revisions reduced the
deceleration in money growth from 11.6
percentage points to 4.8 percentage points.
In another case, the direction of changes
was actually reversed. First-published data
from February 1972 through January 1973
showed M1 grew at an annual rate of 9.9 per-
cent in the first six months and 6.5 percent in
the second six months. The most recent revi-
sion shows a growth of 8.1 percent in the first
six months and 10.5 percent in the second.
Revisions show that instead of slowing in the
second six months, growth in the money
stock was picking up.
Consistent analysis requires adjustments
either to the money stock used in relating
money to economic activity or the first-
published money figures. Otherwise, the in-
terpretation of recent monetary policy is apt
to exaggerate its consequences.
The most important question raised by
the smoothing of money growth rates
through seasonal revisions, however, is
whether data are improved. Every seasonal
adjustment produces a seasonally adjusted
money series. Without an independent
measure of the "true" adjusted money supply
figures, there is no obvious way of deciding
which of these seasonally adjusted series
(and, therefore, the seasonal adjustment fac-
16 Economic Perspectives
tors that produced them) is correct. Revisions
in seasonal adjustments that produce the
smoothest money-stock figures are not
necessarily the best seasonals. As noted
earlier, giving exclusive weight to each
monthly seasonal irregular in determining
the seasonal adjustment factor can purge the
monthly data of all volatility except slight
changes in trend. But too much volatility can
be removed from the series.
The smoothing effects of the seasonal
revisions raise methodological questions in
assessing the seasonal adjustment process.
How can it be said that the seasonal adjust-
ment is known, if it is known only after the un-
adjusted data has been observed? The
problem hangs on the difference between ex-
planation and prediction. Substantial revision
of seasonals can explain away almost all the
volatility in unadjusted money—after the
volatility has been seen. These same revised
seasonal adjustment factors, however, are
much less useful at predicting future money
figures and removing the volatility in the first-
published data.
The size of the seasonal revisions raises
other questions about the seasonal adjust-
ment process. Since calendar months serve as
proxies for contemporaneous economic
forces, the seasonal adjustment of money
assumes that the forces are regular enough to
be represented by calendar months. If,
however, these forces are changing by large
and unpredictable amounts, then it does not
seem reasonable to represent these forces
with seasonals. Instead, it might be better to
leave the influence of the forces in the
seasonally adjusted data and identify these
forces as determinants of the adjusted data.
The influence of policy
One reason seasonal adjustment is more
complicated for money than for most
seasonally sensitive economic series is the
crucial importance of policy in determining
the money stock. Monetary policy can have a
direct and immediate effect on the money
stock. Were this not so, it would make little
sense to interpret monetary policy in terms of
changes in the money stock.
Suppose that for two or three successive
years monetary authorities ran an easy or tight
policy the same month, producing large or
small rates of growth in money. Clearly, the
policy effects should be kept in the seasonally
adjusted money stock series. But unless some
adjustment is made to take account of these
policy effects, the seasonal revision process is
such that policy effects would be treated as a
shift in seasonal factors. They would be
removed from the revised seasonally adjusted
data. Then, in later years when money
figures revert to normal, the first-published
data will be made more volatile by the interim
change in the seasonal. The importance of
policy makes the judgmental contribution to
seasonal adjustment of the money stock an
important factor. Effects of policy could, in
fact, explain the appearance of seasonals in
seasonally adjusted data.
Policy considerations complicate
seasonal adjustment in another way. To a
great extent, the seasonal variation in the
money supply is itself a product of monetary
policy. The seasonal movement is actually a
change in the demand for credit. Before the
Federal Reserve was established, the seasonal
effect was felt primarily on interest rates and
not nearly as much on money. The year-end
increase in credit demands, for example,
resulted largely in higher interest rates. The
increase in money was only slight.
Since the Federal Reserve was establish-
ed, it has acted to dampen the seasonal move-
ment in interest rates. This is done by in-
creasing reserves and money when seasonal
demands for credit rise and reducing them
when credit demands fall. In this way, the
monetary authorities smooth out interest
Federal Reserve Bank of Chicago 17
rates over the year but intensify the seasonal
variation in money.
While it is usually better to leave the
effects of policy in the seasonally adjusted
numbers, policy effects producing a regular
seasonal in money seem different. If year in
and year out, policy produces a seasonal
movement in the money stock, it would seem
better to remove the policy effect from the
seasonally adjusted numbers. The difficulty is
in distinguishing this recurring seasonal
pattern from the ephemeral effects of con-
tracyclical policy that happen to show some
seasonal pattern over a period of years.
Summary
The importance of money and the in-
creasing volatility of the money stock have
focused attention on the seasonal adjustment
of the money stock data. There is clearly a
pronounced seasonal in the money stock. It is
not clear, however, exactly what the pattern is
or how it changes over time.
One important characteristic of the
current method of seasonal adjustment is the
pronounced smoothing in the rate of growth
in the money stock produced by successive
revisions of the seasonal adjustment factors.
This smoothing creates a danger of incon-
sistency in the application of relationships es-
timated with revised data to the analysis of
current monetary policy.
More important, it raises the question of
whether the smoothing process produces
better seasonally adjusted data. The answer
involves serious problems of methodology as
well as economics. Finally, the role of
monetary policy in determining the money
stock makes judgmental modifications of
seasonal adjustment factors important.